Published October 29, 2014

Biometry, Modeling & Statistics

Predicting Maize Phenology: Intercomparison of Functions

for Developmental Response to Temperature
S. Kumudini, F. H. Andrade, K. J. Boote, G. A. Brown, K. A. Dzotsi, G. O. Edmeades, T. Gocken,
M. Goodwin, A. L. Halter, G. L. Hammer, J. L. Hatfield, J. W. Jones, A. R. Kemanian, S.-H. Kim,
J. Kiniry, J. I. Lizaso, C. Nendel, R. L. Nielsen, B. Parent, C. O. Stckle, F. Tardieu,
P. R. Thomison, D. J. Timlin, T. J. Vyn, D. Wallach, H. S. Yang, and M. Tollenaar*
ABSTRACT

Accurate prediction of phenological development in maize (Zea mays L.) is fundamental to determining crop adaptation and yield
potential. A number of thermal functions are used in crop models, but their relative precision in predicting maize development has
not been quantified. The objectives of this study were (i) to evaluate the precision of eight thermal functions, (ii) to assess the effects
of source data on the ability to differentiate among thermal functions, and (iii) to attribute the precision of thermal functions to their
response across various temperature ranges. Data sets used in this study represent >1000 distinct maize hybrids, >50 geographic
locations, and multiple planting dates and years. Thermal functions and calendar days were evaluated and grouped based on their
temperature response and derivation as empirical linear, empirical nonlinear, and process-based functions. Precision in predicting
phase durations from planting to anthesis or silking and from silking to physiological maturity was evaluated. Large data sets
enabled increased differentiation of thermal functions, even when smaller data sets contained orthogonal, multi-location and -year
data. At the highest level of differentiation, precision of thermal functions was in the order calendar days < empirical linear < process
based < empirical nonlinear. Precision was associated with relatively low temperature sensitivity across the 10 to 26C range. In
contrast to other thermal functions, process-based functions were derived using supra-optimal temperatures, and consequently,
they may better represent the developmental response of maize to supra-optimal temperatures. Supra-optimal temperatures could
be more prevalent under future climate-change scenarios, but data sets in this study contained few data in that range.

Modeling crop productivity under climate change has

been a topic of much discussion, and maize is particularly of

significance due to its importance as a global source of food
and feed. Model predictions of crop productivity under climate
change are known to be hindered by uncertainties inherent in
these analyses. Uncertainties can arise from a number of sources
including the choice of general circulation model, the downscaling climate methodology, as well as variation among crop models. In a recent report, Asseng et al. (2013) found that a greater
proportion of the uncertainty in simulating the response of crop
yields to climate change is attributable to differences in crop
models than the uncertainties due to variations among downscaled general circulation models. Crop model differences can
arise from variations in crop parameters or the approaches used
to simulate the underlying biophysical processes. There is a need
to evaluate the accuracy of the processes underlying crop models,
specifically those processes that are most immediately affected by
the environmental factors modified by climate change, such as
temperature, water, and elevated CO2 .

Published in Agron. J. 106:20872097 (2014)

doi:10.2134/agronj14.0200
Available freely online through the author-supported open access option.
Copyright 2014 by the American Society of Agronomy, 5585 Guilford
Road, Madison, WI 53711. All rights reserved. No part of this periodical
may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopying, recording, or any information storage and
retrieval system, without permission in writing from the publisher.

The impact of increased temperature due to climate change

on crop production has been a topic of much interest in the
scientific literature. The Intergovernmental Panel on Climate
Change forecast an increase in global average temperature of
1.8 to 4C from the period 1980 to 1999 to the period 2080
S. Kumudini and M. Tollenaar, The Climate Corp., 110 T.W. Alexander Drive, Research
Triangle Park, NC 27709; F.H. Andrade, Unidad Integrada INTA EEA BalcarceFacultad de Ciencias Agrarias, Univ. Nacional de Mar del Plata, C.C. 276 (7620)
Balcarce, Argentina; K.J. Boote, Dep. of Agronomy, P.O. Box 110500, Univ. of Florida,
Gainesville, FL 32611; G.A. Brown, Breaking Ground, 3501 W. Fillmore St, Chicago, IL
60624; K.A. Dzotsi and J.W. Jones, Dep. of Agricultural and Biological Engineering, P.O.
Box 110570, Univ. of Florida, Gainesville, FL 32611; G.O. Edmeades, 43 Hemans St.,
Cambridge 3432, New Zealand; T. Gocken and M. Goodwin, Monsanto Co., St Louis,
MO; A.L. Halter, Dupont-Pioneer, 5153 E. Simpson Dr., Vincennes, IN 47591; G.L.
Hammer, QAAFI, Univ. of Queensland, Brisbane, QLD 4072, Australia; J.L. Hatfield,
USDAARS, National Lab. for Agriculture and Environment, Ames, IA 50011; A.R.
Kemanian, Dep. of Plant Science, Pennsylvania State Univ., University Park, PA 16802;
S.-H. Kim, School of Environmental and Forest Sciences, College of the Environment,
Univ. of Washington, Seattle, WA 98195; J. Kiniry, USDAARS, 808 E. Blackland Rd.,
Temple, TX 76502; J.I. Lizaso, Dep. Produccin Vegetal: Fitotecnia, Univ. Politcnica
of Madrid, 28040 Madrid, Spain; C. Nendel, Institute of Landscape Systems Analysis,
Leibniz Centre for Agricultural Landscape Research, 15374 Mncheberg, Germany; R.L.
Nielsen and T.J. Vyn, Dep. of Agronomy, Purdue Univ., West Lafayette, IN 47907; B.
Parent and F. Tardieu, INRA, UMR759 Lab. dEcophysiologie des Plantes sous Stress
Environnementaux, Place Viala, F-34060 Montpellier, France; C.O. Stckle, Biological
Systems Engineering, Washington State Univ., Pullman, WA 99164; P.R. Thomison, Dep.
of Horticulture and Crop Science, Ohio State Univ., Columbus, OH 43210; D.J. Timlin,
USDAARS, Crop Systems and Global Change Lab., Beltsville, MD 20705; D. Wallach,
INRA, UMR 1248 Agrosystmes et dveloppement territorial (AGIR), 31326 CastanetTolosan Cedex, France; and H.S. Yang, Dep. Agronomy and Horticulture, Univ. of
Nebraska, Lincoln, NE 68583. Received 14 Apr. 2014. *Corresponding author (matthijs.
tollenaar@climate.com).
Abbreviations: CHU, crop heat units; GTI, general thermal index; GDD, growing
degree days; RM, relative maturity; TLU, thermal leaf units.

A g ro n o my J o u r n a l Vo l u m e 10 6 , I s s u e 6 2 014

2087

T(1 h)
EnzymResp

Tmin, daily minimum temperature; Tmax , daily maximum temperature; T(1 h), estimated from daily Tmin and Tmax using sine-wave temperature distribution during a 24-h period (e.g., Tollenaar et al., 1979); Tmean, daily mean temperature;
T(3 h), estimated from range fractions that describe a sine wave.

Parent and Tardieu (2012)

S(IR/24)
(T + 273)exp{73900/[8.314(T + 273)]}/[1 +
(exp{73900/[8.314(T + 273)]})3.5[1 (T +
273)/306.4)]51,559,240,052

Kim et al. (2012)

S(IR/24)
0.53[(43.7 T)/11.6)(T/32.1)]2.767
T < 0C then T = 0C
T > 43.7C then T = 43.7C
T(1 h)

T < 6C then T = 6C
T > 44.8C then T = 44.8C
Tmin and Tmax
TLU

MAIZSIM

Tollenaar et al. (1979)

(TLUmax + TLUmin)/2

Wilson et al. (1995)

S(IR/8)
T < 0C then T = 0C
T > 44C then T = 44C
T(3 h)
APSIM
Process
based

Tmean
Tmean
GTIveg
GTIrep

0C = < T < 10C: IR = T(10/18C)

18C = < T < 34C: IR = T 8C
34C = < T < 44C: IR = 26C (T 34C)2.6
TLUmin = 0.0997 0.036Tmin + 0.00362Tmin2
0.0000639Tmin3
TLUmax = 0.0997 0.036Tmax + 0.00362Tmax2
0.0000639Tmax3

Stewart et al. (1998)

Stewart et al. (1998)
IR
IR

Brown and Bootsma (1993)

(CHUmax + CHUmin)/2

Tmin < 4.4C then Tmin = 4.4C

Tmax < 10C then Tmax = 10C
Tmin and Tmax
Empirical
nonlinear

CHU

CHUmin = 1.8(Tmin 4.4C)

CHUmax = 3.33(Tmax 10C) 0.084(Tmax 10C)2
0.043177T2 0.000894T3
5.3581 + 0.011178T2

Jones and Kiniry (1986)

S(IR/24)
T < 8C then T = 8C
T > 34C then T = 34C
T(1 h)
CERES-Maize

T 8C

Gilmore and Rogers (1958)

T < 10C then T = 10C
T > 30C then T = 30C
Tmin and Tmax
GDD10,30
Empirical
linear

Reference
Daily rate

C d1
IR
(Tmin + Tmax)/2 10C

Thermal accumulation
Instantaneous rate (IR)
Boundary conditions
Input temperatures
(T)
Model name
Model type

Table 1. Summary of thermal functions, input temperatures (daily), temperature boundary conditions, and instantaneous and daily rates of thermal accumulation. The general thermal index (GTI) model has
separate functions for the pre-and post-flowering periods (GTIveg and GTIrep, respectively).

2088

to 2099 (Intergovernmental Panel on Climate

Change, 2007). Temperature increases can
impact crop production in a number of ways,
but arguably the most important of these is the
impact of temperature on crop phenology. The
importance of phenology for crop productivity
is well understood. The phenology of a crop
will determine its adaptation to a region, its
ability to mature and set grain within a growing
season, and the synchrony of key developmental
phases with ambient environmental conditions
critical for productivity. Consequently,
differences in model prediction of crop
developmental phases can have significant
impacts on the accuracy of the effects of
forecast climate change on crop productivity.
There are a variety of thermal functions
in current crop models that capture the
relationship between temperature and maize
development. These functions vary in a number
of ways, including parameter values, the
quantity and quality of the data from which
they were formulated, the complexity of the
algorithm used to capture the relationship,
and the manner in which the algorithms were
derived (i.e., empirical or process based) (Table
1). The relationship between temperature
and maize development can be derived
either empirically or through a process-based
methodology. For the purposes of this study,
empirically derived functions are defined as
those that are formulated and parameterized
utilizing the same measured phenomenon as the
phenomenon to be predicted (e.g., time from
planting to anthesis). Thermal functions derived
through process-based methodology are defined
as those functions that are parameterized and
derived utilizing information at a lower level of
organization than the phenomenon that is to be
predicted (e.g., enzyme kinetics to predict the
time from planting to anthesis). The thermal
functions studied here are grouped into three
function types that represent simple linear
empirical relationships (Gilmore and Rogers,
1958; Jones and Kiniry, 1986), more complex
empirical nonlinear relationships (Brown,
1969; Stewart et al., 1998), and functions
based on process-based relationships between
temperature and biochemical or organ-level
responses (Tollenaar et al., 1979; Kim et al.,
2012; Parent and Tardieu, 2012). Differences
in cardinal temperatures, as well as the
relative rates of development within specific
temperature ranges, are seen across the thermal
functions studied (Table 1).
Despite the importance of thermal functions
in the prediction of crop developmental stages
and the variation in the currently utilized
thermal functions, there is no clear indication

Agronomy Journal Volume 106, Issue 6 2014

of the relative precision of these functions in simulating crop

developmental phases. Fundamental differences in the accuracy
of these algorithms are largely unknown, partially due to the
use of genotype coefficients that fit the genotype to a set of
data. Therefore, the evaluation of thermal functions for the
accurate simulation of maize development (while limiting
potential biases imposed by the use of genotype coefficients) is
an important inquiry.
In assessing the impact of climate change on crop productivity,
assumptions are generally made about the hybrids adapted for
a region or the date on which the hybrid was probably planted.
In some applications of crop models, such as gridded models,
instead of specific hybrids an average hybrid is used to capture
the average response of hybrids adapted to the geographic region.
To indicate the geographic region of adaptation, commercial
hybrids are grouped by comparative relative maturity (Lauer,
1998), or simply relative maturity (RM). The determination of
the developmental response of RM groups is therefore of value to
modelers to simulate the average response of hybrids adapted
to specific geographies. Variation in the precision of the thermal
functions for different RM groups may indicate that some
thermal functions are better able to capture the temperature
variation extent in the geographies to which the RM groups are
adapted or, alternatively, that RM group hybrids have different
temperature responses. Therefore, the relationship between
RM groups and thermal function precision is an important
consideration. An early vs. a delayed planting date may also
impact the precision of thermal function in much the same way
as RM group, again calling for consideration of how the precision
of thermal functions may vary by the temperature variations that
can result as a consequence of a change in planting date.
The overall goal of this study was to bring together researchers
to evaluate a number of thermal functions for their ability to
predict maize phenology across a range of source data that vary
in the number of hybrids, geographies, and environmental
conditions tested. The specific objectives of this study were (i)
to assess the effect of source data on the ability to differentiate
among thermal functions, (ii) to quantify the relative precision
of eight thermal functions in predicting maize phenological
development, and (iii) to attribute the precision of thermal
functions to their response across various temperature ranges.
MATERIALS AND METHODS
Thermal Functions
Eight thermal functions were selected for evaluation and
classified according to their derivation type (i.e., empirical or
process based) and thermal response. An empirical function is
defined herein as a function for which the level of organization
of its derivation and parameterization is the same as that of its
outcome, e.g., the planting to anthesis interval of a field-grown
maize canopy. In contrast, a process-based function is defined
as a function that is derived and parameterized at a lower level
of organization than that of its outcome, e.g., the derivation
and parameterization of a function at the enzyme-kinetics level,
whereas the outcome of the function is the planting to anthesis
interval of a field-grown maize canopy. For the purposes of this
study, the APSIM, TLU, MAIZSIM, and EnzymResp functions
(defined below) were classified as process-based functions, and the
GDD10,30, CERES-Maize, CHU, and GTI functions (defined

below) were classified as empirical functions. The empirical

functions were subdivided into linear (GDD10,30 and CERESMaize) and nonlinear (CHU and GTI) functions, based on
their temperature response at a constant temperature during a
24-h period. The thermal functions evaluated in this study are
summarized in Table 1 and defined as follows:
1. GDD10,30: Growing degree days (GDD) with a base temperature of 10C and an optimum temperature at 30C is
a thermal function developed by Gilmore and Rogers (1958)
using 10 hybrids and 10 inbred lines grown at College
Station, TX, at five planting dates in 1956. They calculated
heat unit accumulation by subtracting 10C from the daily
minimum and maximum temperatures before calculating
daily means, using six combinations of base (none or 10C)
and optimum (none, 30, or 32.2C) temperatures. The
method with the highest precision in their study was the
classic 10/30C cutoff method.
2. CERES-Maize (GDD8,34): The CERES-Maize model utilizes GDD8,34, with temperatures at hourly time steps estimated from the minimum and maximum temperatures. The
GDD8,34 is a linear function similar to GDD10,30, but with
base and optimum temperatures of 8 and 34C, respectively.
The original thermal function for CERES-Maize described
by Jones and Kiniry (1986) used eight 3-h, third-order polynomial interpolations, but in the current version, thermal
accumulation is calculated from temperatures during 24 1-h
periods (J.T. Ritchie, personal communication, 1998), which
are estimated using a sine wave between the minimum and
the maximum daily temperatures (cf., Tollenaar et al., 1979).
The Hybrid-Maize (Yang et al., 2006) and IXIM (Lizaso et
al., 2011) models also use GDD8,34, with eight 3-h and 24
1-h temperatures, respectively, estimated from minimum
and maximum temperatures.
3. APSIM: The APSIM model uses a single multilinear function for thermal accumulation that reflects the biological
response of development across the 0 to 44C range in temperatures encountered in most environments in which maize
is cultivated (Wilson et al., 1995). This temperature-response
function has 0C minimum, 34C optimum, and 44C
ceiling temperatures. Daily thermal accumulation was calculated from eight 3-h, third-order polynomial interpolations
between the minimum and maximum daily temperatures.
4. Thermal leaf units (TLU). Analogous to accumulated growing degrees days in GDD10,30 (C d), TLU is accumulated
thermal leaf units (L, leaf tips) by conversion of temperature
to leaf (tip) units through the temperature-dependent rate
of leaf appearance (L C1 d1 C d = L), where the rate
of leaf appearance (RLA) is the inverse of the phylochron.
The relationship between the rate of leaf-tip appearance and
temperature (Tollenaar et al., 1979) represents a biologically
meaningful temperature response. This response is similar
to other growth and development processes in maize such
as seedling growth (Lehenbauer, 1914), radical elongation
(Blacklow, 1972), and leaf elongation (Parent and Tardieu,
2012). The temperature-compensated rate of leaf-tip appearance is fairly constant until the emergence of the topmost
leaf (Tollenaar et al., 1979, 1984). The number of visible
leaf tips per plant is assumed to be equal to the accumulated
TLU at any time between planting and the appearance of

Agronomy Journal Volume 106, Issue 6 2014

2089

the topmost leaf if the rate of leaf appearance is influenced

by temperature only. Based on this assumption, TLU accumulation between planting and anthesis is equal to the total
leaf number plus the TLU accumulation during the interval
between the appearance of the topmost leaf and anthesis.
Leaf-ligulae appearance has also been used to quantify (e.g.,
Warrington and Kanemasu, 1983a) and to model (e.g.,
Parent et al., 2010) the rate of leaf appearance and maize development, among others, because of the wide use of V stages
(Abendroth et al., 2011) to describe phenological development in maize. In contrast to the temperature-compensated
rate of leaf-tip appearance, the temperature-compensated
rate of leaf-ligule appearance varies widely for the planting to
flowering interval. The ratio of the rate of leaf-tip appearance
to the rate of leaf-ligule appearance is >1 during the first
eight to 12 leaves, and it is <1 during the second half of the
pre-silking period (Muldoon et al., 1984). The variability of
the temperature-compensated rate of leaf-ligule appearance
across the pre-silking phase diminishes its suitability for use
in modeling the rate of maize development. Analogous to accumulated growing degrees days in GDD10,30 (C d), TLU,
(L C1 d1 C d = L) are accumulated by the conversion
of temperature to leaf tips through the RLA function.
5. MAIZSIM (b function): A b function was introduced by
Yin et al. (1995) that described the temperature response
of crop development. This function describes a smooth
non-symmetric response to temperature using five parameters: three cardinal temperatures (the base temperature
[Tbase], the optimum temperature [Topt], and the maximum temperature [Tceil]), a parameter for the maximum
rate of development at Topt (Rmax), and a parameter that
describes the curvature of the relationship. All parameters
except the parameter that describes the curvature are biologically meaningful. Yan and Hunt (1999) simplified the
Yin et al. (1995) function by eliminating the variable describing the shape of the curve and by setting Tbase to zero,
resulting in the following equation with three parameters:
T

T -T T opt

R (T ) = Rmax ceil
Tceil -Topt Topt

/(Tceil -Topt )

[1]

where R(T) is the rate of development as a function of temperature. Yan and Hunt (1999) showed that this simplified b
function was highly predictive when model parameters that
were estimated from six constant temperatures were used
to predict the rates of 16 varying daynight temperature regimes in the Tollenaar et al. (1979) data set. The MAIZSIM
model (Kim et al., 2012) uses Eq. [1] for predicting the leaf
appearance rate and assigned the following parameter values:
Topt = 31.2C, Tceil = 43.7C, and Rmax = 0.53 d1.
6. Enzymatic response (EnzymResp): Using an equation for
describing the temperature response of enzyme activities
( Johnson et al., 1942), Parent et al. (2010) showed that
processes linked to plant growth and development showed
coordinated temperature responses and followed common
laws within a plant species. Interestingly, they showed that
there was no coordination of temperature responses linked
to plant metabolism and leaf photosynthesis. In a meta2090

analysis, Parent and Tardieu (2012) utilized an equation

to show a common response among maize genotypes from
high-latitude, temperate, and tropical regions across a wide
range in temperatures:

R(T ) =

AT exp -DH A RT

a(1-T T0 )
1 + exp -DH A RT

[2]

The function has two parameters: T0 = 306.4K ( = 33.3C)

and a = 3.5 (Parent and Tardieu, 2012). The parameter
DHA = 73900 J mol1 is deduced from the other two. and
A is a size factor without units (51,559,240,052).
7. Crop heat units (CHU): Crop heat units have been used in
Canada to account for temperature effects on phenology in
maize since the mid-1960s. Crop heat units are estimated
using separate night and day temperature functions (Table 1).
The former is a linear function with a minimum of 4.4C and
the latter is a quadratic function with a minimum at 10C
and an optimum near 30C (Brown, 1969). The quadratic
function was developed by Brown (1960) from data collected
on soybean [Glycine max (L.) Merr.] grown in environmentcontrolled studies (Van Schaik and Probst, 1958).
8. General thermal index (GTI): The GTI model is based on a
large data set of maize hybrids grown during 4 yr at 19 locations across a 9interval in latitude across the North American
Corn Belt, in which the durations from planting to silking
and from silking to maturity were related to air temperature
(Stewart et al., 1998). The GTI model is unique in that (i)
it consists of two polynomial responses to temperature, one
for the period from planting to silking and one for the period
from silking to maturity, and (ii) the temperature response
during the post-silking period was flat relative to that during
the pre-silking period, particularly for temperatures <20C
(cf., Stewart et al., 1998, Fig. 2 and 5). Although the relative insensitivity of the duration of the grain-filling period in
maize to temperature has been reported previously (Shaw and
Thom, 1951; Brown, 1977), none of the current thermal models other than GTI account for differences in the temperature
response between the pre- and post-flowering periods.
Thermal functions can have a range of responses across
temperatures and, consequently, the precision of the thermal
functions will probably be related to their differential response
across the temperature ranges to which they are exposed.
Most figures of thermal functions in the literature graph the
developmental response against a specific temperature value,
which is useful for evaluating the function at the hour when
the plant experienced the specific temperature. However,
due to the diurnal range in temperatures to which plants are
exposed daily, an hour snap shot is a limited perspective of the
plants daily developmental response. Given that a locations
diurnal temperature variation is relatively stable, the response
of thermal functions for a given diurnal range can be computed
and graphed against a range of mean temperatures. To better
capture how the thermal functions in this study respond to
daily ambient conditions, the response of the thermal functions
to mean temperature in the 0 to 40C range was calculated
assuming a diurnal temperature range of 12C (i.e., the
difference between Tmax and Tmin values). The 12C diurnal
Agronomy Journal Volume 106, Issue 6 2014

temperature range value was selected because this was the

approximate mean diurnal temperature range in the data sets
utilized in the current study. To facilitate a comparison of the
relative response, each function was normalized for the rate at
its optimum temperature (Fig. 1).
Data Sources and Analyses
The precision of the thermal functions was evaluated using
data from three different sources:
1. North America: Data were collected in field trials performed
by Monsanto Co. that resulted in two different data sets.
First, data were collected in trials that were performed from
2007 to 2011 at 43 locations across the North American
Corn Belt. A total of 118 commercial DeKalb maize hybrids
were tested that ranged from 76 to 119 RM. The hybrids
were grown in their area of adaptation, and the number of
locations in each of five RM classes was 13 for RM 76 to 85,
27 for RM 86 to 95, 26 for RM 96 to 105, 28 for RM 106
to 115, and 14 for RM 116 to 119 (i.e., the total number of
locations is 43 because hybrids of more than one RM class
were tested in several locations). The number of hybrids and
locations tested varied each year. The data set consisted of
1375 combinations of hybrids locations years. Dates
of planting and 50% anthesis were measured in this data
set. The second data set was collected in trials that were
performed from 2007 to 2012 at a location near DeKalb,
IL. This location was not included in the 43-location North
American data set. At this location, dates of black layer
formation (i.e., physiological maturity), anthesis, and silking
were recorded on 3129 hybrids years, and the hybrids
tested included all 118 hybrids grown at the other 43 locations. The hybrids in this data set ranged in RM from 76 to
119. At all locations, maize hybrids were grown in 6.1-mlong, four-row plots using standard agronomic practices.
2. IndianaOhio: This data set consisted of 108 entries. Details
of this study have been previously reported by Nielsen et al.
(2002). In short, three Pioneer maize hybrids (CRM 106,
111, and 115) were grown at three planting dates, ranging
from 22 April to 17 June, each at four locations in Ohio and
Indiana from 1991 to 1994. Note that Pioneer uses CRM
(comparative relative maturity) to designate hybrid RM and
that hybrid RM ratings may vary somewhat among different seed companies (Lauer, 1998). Dates of 50% silking and
50% black layer were recorded.
3. Argentina: This data set consists of information collected
on various maize hybrids ranging from RM 92 and 127 that
were grown between 1988 and 2013 near Balcarce, Argentina. Data included the results of a published study (Cirilo
and Andrade, 1994a, 1994b) and several unpublished
studies. Some of the studies had multiple planting dates that
ranged from September to January. The experimental area
was irrigated when deemed necessary. Dates of 50% silking
and 50% black layer were recorded.
Weather data were collected for each location-year
in the three data sets. Records of daily maximum and
minimum air temperatures were obtained either from the
National Climate Data Center (U.S. weather stations,
http://www.ncdc.noaa.gov/cdo-web/search) or the Government
of Canada Climate Service (Canadian weather stations,

Fig. 1. Relationships between mean daily temperatures with a 12C diurnal

range normalized for rate of development at the optimum temperature
specific to each function: (a) growing degree days (GDD10,30), CERESMaize, and crop heat units (CHU) functions, (b) APSIM, thermal leaf
units (TLU), MAIZSIM, EnzymResp, and CHU functions, and (c) general
thermal index (GTI) during the preflowering period (GTIV), GTI during
the post-flowering period (GTIR), and CHU functions.

Agronomy Journal Volume 106, Issue 6 2014

2091

http://climate.weather.gc.ca/index_e.html#access) for the

location-years in the North American and IndianaOhio data
sets. All weather service stations were, on average, within 20 km
of the experimental site.
Field data were analyzed for the pre-flowering and postflowering periods. The pre-flowering period was defined as the
period from the first day after planting until, and including,
anthesis (i.e., the first day at which 50% of plants shed pollen)
or silking (i.e., the first day at which silks had emerged from
the topmost ear shoot in 50% of plants), and the post-silking
period was defined as the first day after silking until, and
including, black layer (the first day on which 50% of plants
have grain with a black layer in the center portion of the ear).
Whenever possible, the planting to anthesis interval rather
than the planting to silking interval was analyzed because
the anthesis to silking interval is highly affected by abiotic
stresses (Edmeades et al., 2000) and, consequently, the use of
the silking date will introduce another potential unaccounted
source of variability in the data.
Daily thermal accumulation was estimated using the daily
mean temperature (GTI), minimum and maximum temperatures
(GDD10,30 and CHU), 1-h temperatures (CERES-Maize,
TLU, MAIZSIM, and EnzymResp), and 3-h temperatures
(APSIM). The 1-h temperatures were estimated using a sine wave
between the minimum and the maximum daily temperature
(e.g., Tollenaar et al., 1979). The 3-h temperatures in the APSIM
function were estimated using range fractions for each of the
eight time periods within a 24-h day that describe a sine wave.
The CV (coefficient of variability) was used to quantify
the precision of the thermal functions. The mean thermal
accumulation (and days), standard deviation, and CV were
calculated from all observations (i.e., hybrid-location-year)
within an RM class for the North American data set and within
a planting date for the IndianaOhio and Argentinean data sets
for the planting to anthesis (or silking) interval and the silking
to physiological maturity interval. Statistical analyses of the CVs
were performed using the Glimmix procedure for generalized
linear modeling assuming a log link and g distribution in SAS,
Version 9.3, where RM classes or planting dates were replications
for the statistical comparisons among thermal functions,
and thermal functions were replications for the statistical
comparisons among either RM classes or planting dates.
Differences in the precision of the thermal functions across RM
groups may be due to genetic or environmental factors associated
with the regions to which the RM groups are adapted. One
method for deconstructing these factors is to identify differences
in cardinal temperatures among RM classes. Since the MAIZSIM
function explicitly quantifies Topt and Tceil values, optimization
of this function for each of the five RM classes can identify
these cardinal temperatures of each RM class. The optimization
procedure of the b function was performed by varying the values
of Topt and Tceil (keeping T base unchanged at 0C) and selection
based on the lowest CV. The optimization of the b function was
done using the optimum function of the R statistical language
(R Development Core Team, 2013) with the NelderMead
algorithm for both entries in each of the five RM classes of the
43-location North American data set and for all RM classes
combined (bOPT). This is a general purpose routine for function
error minimization. Several of the minimization calculations were
2092

run for two different starting values, and the results were identical
in the two cases. This exercise confirmed that the algorithm was
converging to a global rather than to a local optimum.
In addition, for each of the five RM class in the 43-location
North America data set, the hybrid-location-years of the data
set were subdivided equally into three groups representing 33%
low, 33% medium, and 33% high means of daily minimum
temperature and of daily maximum temperature, resulting
in 15 mean-minimum and 15 mean-maximum temperature
groups (i.e., 5 RM classes 3 groups within each RM class).
This analysis is informative of thermal function sensitivity
variation by temperature. The thermal functions tested
vary in T base, Topt, and Tmax values. Differences in cardinal
temperature may render some thermal functions more sensitive
to specific temperatures than others, thus impacting their
relative precision in simulating development.
RESULTS AND DISCUSSION
Impact of Data Source on Thermal
Function Evaluation
Data source is an important variable when evaluating the
precision of thermal functions. The data sets evaluated in this
study each had limitations. The IndianaOhio data set was
comprised of a multilocation, multiyear study with orthogonal
treatments, but contained only a relatively small number of
hybrids and environments (i.e., location-years). In contrast, the
North American data sets consisted of a large number of hybrids
and environments, but the treatments were not orthogonal.
Different hybrids were grown in each location-year, resulting
in a different number of observations for each RM class in the
43-location data set (Tables 2 and 3). In addition, the silking
to black layer interval was evaluated only at a single geographic
location near DeKalb, IL, for multiple years (Table 3). The
Argentinean data set was not orthogonal and contained a
relatively small number of environments (years).
Empirical nonlinear functions were more precise than either
empirical linear or process-based functions in the North American
and IndianaOhio data sets, but the precision of the thermal
functions did not differ in the Argentinean data set (Tables 26).
Differences among thermal functions and calendar days were
generally smaller in the IndianaOhio than the North American
data set. The precision of the thermal functions for the planting
to anthesis phase in the North American data set were calendar
days < empirical linear < process based < empirical nonlinear,
whereas in the IndianaOhio data set they were calendar days <
empirical linear = process based < empirical nonlinear. In the
Argentinean data set, calendar days were less precise than the
thermal functions, but the thermal functions could not be
differentiated. Therefore, large data sets have improved ability to
differentiate thermal function performance, even when they may
have limitations such as lack of orthogonality. Ideally, large and
orthogonal data sets would be preferred.
Relative Precision of Thermal Functions
Within the large data sets that allowed the greatest
differentiation among the thermal functions, the empirical
nonlinear functions proved to be more precise than the other
functions tested (Tables 26). Thermal functions were generally
more precise than calendar days, although differences were
Agronomy Journal Volume 106, Issue 6 2014

Table 2. Coefficients of variation of the number of days and thermal accumulation during the planting to anthesis interval determined by eight functions for commercial hybrids across five relative maturity (RM) groups grown at 43 locations in the Corn Belt from 2007 to 2011 (North American
data set). The number of observations (n) is the sum of all hybrids grown in each location-year for a RM class.
Coefficient of variation
RM
7685
8695
96105
106115
116122
Mean
All

n
121
204
342
560
148
1375

Days

GDD10,30

CERES

MAIZSIM

TLU

EnzymResp

APSIM

GTI

CHU

%
10.3
9.5
9.3
9.0
8.4
9.0
7.8
7.6
7.0
9.9
9.7
9.5
9.2
8.5
9.2
7.7
7.6
6.8
9.3
6.4
6.5
6.2
6.0
6.3
6.0
5.7
5.6
10.4
6.4
6.9
6.0
5.3
6.2
5.4
4.7
4.5
10.3
6.3
6.9
5.9
5.3
6.0
5.2
4.5
4.6
10.1 e
7.7 d
7.8 d
7.3 cd
6.7 c
7.3 d
6.4 bc
6.0 ab
5.7 a
10.7
10.1
10.2
9.5
8.7
9.6
7.9
7.5
6.7

Means within a row followed by the same letter are not significantly different at the 0.05 level of probability.

Table 3. Coefficients of variation of the number of days and thermal accumulation during the silking to black layer interval determined by eight thermal
functions for commercial hybrids across five relative maturity (RM) classes grown at a single location in the U.S. Corn Belt from 2007 to 2012 (North
American data set). The number of observations (n) is the sum of all hybrids grown in each location-year for a RM class.
Coefficient of variation
RM
7685
8695
96105
106115
116122
Mean
All

n
182
413
771
1477
286
3129

Days

GDD10,30

CERES

MAIZSIM

TLU

EnzymResp

APSIM

GTI

CHU

%
9.0
8.2
8.0
7.6
7.2
8.2
7.1
6.2
6.0
9.4
9.3
9.0
8.6
8.2
9.3
8.2
7.4
6.8
8.5
9.4
9.3
8.7
8.0
9.5
8.3
7.1
6.0
9.9
9.6
9.4
8.9
8.2
9.6
8.1
7.1
5.8
9.6
11.8
11.5
11.0
10.3
11.3
9.6
7.7
7.4
9.3 de
9.7 e
9.5 de
9.0 d
8.4 c
9.6 e
8.3 c
7.1 b
6.4 a
12.4
10.2
10.1
9.7
9.1
10.3
9.4
9.0
7.7

Means within a row followed by the same letter are not significantly different at the 0.05 level of probability.

Table 4. Coefficients of variation of the number of days and thermal accumulation during the planting to silking and silking to black layer intervals determined by eight thermal functions for three maize hybrids planted at three dates from April to June during 1990 to 1994 at four locations in Ohio
and Indiana (Nielsen et al., 2002); the total number of observations is 108. Data were analyzed both separately for the early, medium, and late planting
dates and together for all planting dates.
Coefficient of variation
Planting date

Days

Early
Medium
Late
Mean
All

11.6
9.1
5.6
8.8e
12.9

Early
Medium
Late
Mean
All

9.5
11.5
8.6
9.9bc
10.3

GDD10,30

CERES

MAIZSIM

TLU

EnzymResp

APSIM

GTI

CHU

%
Planting to silking
4.4
5.2
4.7
3.6
4.5
5.9
4.4
3.5
4.7
5.4
4.9
3.9
4.9
6.1
4.5
3.8
5.7
6.4
6.1
5.2
6.1
7.2
5.7
4.7
4.9abc
5.7cd
5.3bcd
4.2ab
5.2bcd
6.4d
4.8abc
4.0a
5.2
5.8
5.5
4.7
5.5
7.0
5.4
5.0
Silking to black layer
8.4
9.2
9.0
7.9
8.7
10.1
6.8
8.0
9.0
10.2
10.0
8.6
9.4
11.0
7.6
8.9
11.3
12.7
12.5
10.7
11.8
14.0
8.6
10.5
9.6bc
10.7cd
10.5cd
9.0b
10.0bc
11.7d
7.7a
9.1b
11.8
13.1
12.8
11.0
12.0
12.8
7.9
10.5

Means within row followed by the same letter are not significantly different at 0.05 level of probability.

smaller during the period between silking and black layer than
during the planting to anthesis interval. The precision of the
thermal functions for both the pre-flowering and post-flowering
period in the North American data set was empirical linear <
process based < empirical nonlinear (Table 6). The CHU and
GTI functions were the most precise for the prediction of both
the pre-anthesis and post-silking phases in the North American
data set (Tables 2 and 3). In the IndianaOhio data set, the GTI
function was the only thermal function that was more precise
than calendar days for the silking to black layer period (Table 4).
The post-flowering GTI thermal function had a very different

thermal response than any other thermal function tested (Fig.

1). The higher precision of GTI for the post-flowering period
was associated with a relatively small difference in thermal
accumulation from silking to maturity between the early and
late planting dates for GTI. Nielsen et al. (2002) showed a 17%
difference in GDD10,30 accumulation during the post-flowering
period between early and late planting, whereas the difference
in GTI accumulation was only 5% (data not shown). The mean
minimum daily temperatures during the silking to black layer
interval decreased from 14.3 to 10.7C for the early and late
planting dates, respectively.

Agronomy Journal Volume 106, Issue 6 2014

2093

Table 5. Coefficients of variation of number of days and thermal accumulation by eight functions during the planting-silking and silking-black layer intervals of maize hybrids planted from September to January between 1989 and 2012 at locations near Balcarce (Argentina). Data were analyzed both
separately for four planting-date periods and together for all planting dates. The number of observations (n) is the sum of all hybrids grown in each
location-year for a RM class.
Coefficient of variation
Planting date

Sept.
Oct.
Nov.
Dec.Jan.
Mean
All
Sept.
Oct.
Nov.
Dec.Jan.
Mean
All

6
41
13
16
76
6
41
13
16
76

Days

GDD10,30

CERES

MAIZSIM

TLU

EnzymResp

APSIM

GTI

CHU

%
Planting to silking
5.4
8.2
7.8
7.7
7.6
7.6
6.7
7.0
6.9
6.7
6.2
6.5
6.2
5.6
6.4
6.5
6.3
5.5
9.0
8.5
8.7
8.5
8.3
8.7
8.6
8.4
8.1
12.5
9.9
10.0
9.9
10.0
9.9
10.2
10.2
10.8
8.4 a
8.2 a
8.2 a
8.1 a
7.9 a
8.2 a
8.0 a
8.0 a
7.8 a
18.5
8.8
9.1
9.0
9.0
9.2
10.3
10.1
10.7
Silking to black layer
6.7
6.0
5.0
5.4
5.7
5.6
5.1
5.4
5.4
12.5
12.6
13.0
12.7
12.1
12.8
12.9
12.7
11.9
18.7
11.0
10.3
10.8
11.9
11.0
11.8
13.6
13.3
11.0
8.6
8.5
8.4
8.4
8.1
7.6
8.2
8.7
12.2 b
9.6 a
9.2 a
9.3 a
9.5 a
9.4 a
9.4 a
10.0 a
9.9 a
13.0
17.0
17.5
16.8
15.7
17.0
15.7
14.1
14.0

Means within row followed by the same letter are not significantly different at 0.05 level of probability.

Table 6. Mean coefficients of variation of calendar days and thermal accumulation determined by empirical-linear functions (GDD10,30 and CERESMaize), process-based functions (APSIM, MAIZSIM, EnzymResp, and TLU), and empirical-nonlinear functions (GTI and CHU) during the planting to
anthesis or silking and the silking to black layer intervals in the 43-location, 5-yr and the one-location, 6-yr North American data sets (Tables 2 and 3),
the IndianaOhio data set (Table 4), and the Argentinean data set (Table 5).

Phase of development
Planting to anthesis or silking

Silking to black layer

Data set
North America
IndianaOhio
Argentina
North America
IndianaOhio
Argentina

Coefficient of variation
Thermal function
Calendar days
Empirical linear
Process-based
Empirical nonlinear
%
10.1 d
7.7 c
6.9 b
5.9 a
8.8 c
5.3 b
5.3 b
4.4 a
8.4
8.2
8.0
7.9
9.3 c
9.6 c
8.8 b
6.8 a
9.9 ab
10.1 b
10.3 b
8.4 a
12.2 b
9.4 a
9.4 a
9.8 a

Means within a row followed by the same letter are not significantly different at the 0.05 level of probability.

Relationship between Temperature Response

and Precision of Thermal Functions
The precision of the thermal functions is associated with
their response across a wide range of temperatures. One
means of understanding the precision of thermal functions
is to compare their normalized response across a range of
temperatures in relation to that of CHU, which has been
shown to be more precise in the North American data set
(Fig. 1). The results of the normalized responses to mean
temperatures (based on a 12C diurnal range) show that: (i)
the empirical functions (GDD10,30 and CERES) have a linearlike response between mean temperatures of 4 and 36C (Fig.
1a) (the generally depicted temperature response of GDD10,30
in the literature is a straight line between 10 and 30C,
but this is true only when there is no diurnal temperature
variation or if the maximum and minimum daily temperatures
are >10C and <30C); (ii) the process-based functions (TLU,
APSIM, MAIZSIM, and EnzymResp) have a temperature
response similar to each other in the 15 to 45C temperature
range, with small differences in the lower temperature range
(Fig. 1b); and (iii) the nonlinear empirical functions are
similar to each other across 0 to 30C for the pre-silking
phase (Fig. 1c). The post-silking GTI function has a very
2094

different temperature response that is relatively insensitive at

temperatures <23C (Stewart et al., 1998).
In this study, the nonlinear empirical functions were the
most precise group of thermal functions tested, and differences
in precision were associated with response in the 10 to 26C
mean diurnal temperature range (Fig. 1). Unlike linear empirical
functions, the shape of process-based functions and the nonlinear
empirical function are similar (Fig. 1). These two groups of
thermal functions varied somewhat in their base temperatures
(Fig. 1b and 1c), but there was no consistent relationship between
the base temperatures and the precision of the function (Tables
25). All three groups differ in the supra-optimal temperature
range. However, differences in precision cannot be attributed to
the developmental responses in the supra-optimal temperature
range because none of the data sets used in this study contained
mean daily temperatures >30C. Differences in precision among
thermal functions were associated with their relative temperature
response in the 10 to 26C mean diurnal temperature range (Fig.
1). The CHU function had higher developmental rates across
the 10 to 26C range and was the most precise in predicting
phenology. At 20C, for instance, the rate of development
of CHU is 21% greater than that of TLU, and the rate of
development of TLU is 21% greater than that of GDD10,30 (Fig.
Agronomy Journal Volume 106, Issue 6 2014

Table 7. Mean coefficient of variations (CV) of thermal accumulation determined by eight functions during the planting to anthesis and silking to black
layer intervals, and mean daily minimum and maximum temperatures of locations for five relative maturity (RM) groups. The planting-to-anthesis
interval comprised observations across 43 Corn-Belt locations during 2007 to 2011 (Table 1) and the silking to black layer interval comprised > 3100
observations at one Corn-Belt location from 2007 to 2012 (North American data sets).

RM

CV

7685
8695
96105
106115
116122

8.5 c
8.5 c
6.1 b
5.7 a
5.6 a

Planting to anthesis
Daily temperature
Min.
Max.

Silking to black layer

Daily temperature
Min.
Max.

CV
%

C
11.7
23.1
12.1
23.9
13.0
25.4
14.3
26.3
15.7
27.5

C
17.1
28.4
16.9
27.9
16.6
27.7
15.6
26.7
15.1
26.3

7.3 a
8.4 b
8.3 b
8.3 b
10.1 c

Means within a column followed by the same letter are not significantly different at the 0.05 level of probability.

1), while precision of the three functions in the North American

data sets was in the order CHU > TLU > GDD10,30 (Tables 2
and 3). Thermal functions with higher normalized developmental
rates across a wide range of temperatures (e.g., nonlinear empirical
functions) have lower temperature sensitivity.
The precision of the thermal functions varied with RM class
in the North American data set, and the variation in precision
was associated with air temperature. The precision of the thermal
functions increased from low- to high-RM classes during
the planting to anthesis interval (Table 7) in the 43-location
North America data set. This relationship between RM class
and thermal function precision could be due to either genetics
or environment. The association with genetics was dismissed
because, when all RM hybrids were grown at a single location
(i.e., the DeKalb, IL location), the trend in RM precision was no
longer observed, i.e., there was either an opposite trend during
the silking to black layer period (Table 7) or no trend during
the planting to anthesis interval (data not shown). The effects
of temperature on the relative precision of thermal methods
across the RM groups in the North American data set was
examined in more detail by subdividing each of the five RM
groups of environments into three mean-minimum-temperature
environments (Fig. 2). In this analysis, the change in CV of
each thermal function across the 15 temperatureRM groups

was 0.55% C1 for GDD10,30, 0.49% C1 for TLU, and

0.41% C1 for CHU, but these slopes were not significantly
different (data not shown). The CVs in some of the temperature
environments deviated substantially from the regression (Fig. 2),
and variation among the three thermal functions within these
environments was small when the deviation from the regression
was negative (e.g., the 10.2, 12.3, and 13.9C environments) and
was large when the deviation from the regression was positive
(e.g., the 10.4, 12.0, and 13.4C environments).
When the b function was optimized for five RM groups, the
values of Topt and Tceil of the early RM classes (Table 8) were
much lower than reported values from controlled-environment
studies (e.g., Parent et al., 2010; Sanchez et al., 2014). The low
Topt and Tceil values were probably not due to genetic differences
among RM groups. Low Topt (21.2 and 22.3C) has been
reported for the temperature-dependent duration of the period
from sowing to tassel initiation for two highland tropical inbred
lines in controlled-environment studies performed by Ellis et al.
(1992) and reported by Yin et al. (1995), but neither the results
of Tollenaar et al. (1979), which included North American
hybrids of <85 RM, nor the results of Parent and Tardieu (2012),
which included very early European hybrids and inbred lines,
showed Topt and Tceil values that were lower than those reported
in studies of temperature-dependent rates of growth and
development in maize (cf., Parent and Tardieu, 2012).
The relatively high precision of temperature-insensitive
functions, the decline in CV with increase in temperature
among RM groups, and low cardinal temperatures in the
optimized b function of early RM groups all support the
contention that maize phenology is influenced by factors other
than air temperature per se. Factors other than air temperature
Table 8. Coefficients of variation (CV) for the planting to anthesis interval of thermal functions that utilize different forms of the b function:
the MAIZSIM b function, and different b functions modified in order to
optimize CV by changing Topt and Tceil within each of the five relative
maturity (RM) classes (Tbase is held constant at 0C).
Coefficient of variation
bOPT

RM

Fig. 2. Relationship between mean minimum temperature and CVs

of the growing degree day (GDD10,30), thermal leaf units (TLU), and
crop heat units (CHU) functions for 12 environments (location-years).
Environments were grouped into the 33% highest, 33% medium, and
33% lowest mean minimum temperature environments in each of the
five relative maturity classes depicted in Table 7.

d
7685
8695
96105
106115
116122

MAIZSIM

%
6.2 (23.0, 32.4)
9.0 (32.1, 43.7)
6.3 (25.7, 39.6)
9.2 (32.1, 43.7)
5.4 (26.2, 35.9)
6.2 (32.1, 43.7)
4.3 (27.6, 37.8)
6.0 (32.1, 43.7)
4.3 (30.3, 44.9)
5.9 (32.1, 43.7)

Topt , Tceil (C) in parentheses.

Agronomy Journal Volume 106, Issue 6 2014

2095

that can influence phenology include soil and apex temperature

(Vinocur and Ritchie, 2001), photoperiod (e.g., Kiniry et al.,
1983; Tollenaar and Hunter, 1983; Warrington and Kanemasu,
1983b), incident solar radiation (Birch et al., 1998, Tollenaar,
1999), and periods of severe stress (e.g., McCullough et al.,
1994). For instance, apex temperature is influenced by plant
transpiration and soil temperature before the eight-leaf-tip stage
and the apex-air temperature differential appears to be greatest at
low temperatures (Vinocur and Ritchie, 2001; Birch et al., 1998).
The apex-air temperature differential may have played a role
in the higher precision of less temperature-sensitive functions,
the increase in precision with temperature among RM groups,
and the low Topt and Tceil values for the early RM groups in
the MAIZSIM optimization. The lower precision of processbased functions may be due to the fact that these data sets were
generated under controlled-environment conditions where many
of the factors other than air temperature were held constant and
where the apex-air temperature differential may have been less
than under field conditions.
Precision of Thermal Functions under
Supra-Optimal Temperatures
Climate change will probably result in an increase in the number
of days in the growing season with elevated temperatures, and
therefore an accurate estimate of the impact of supra-optimal
temperatures on maize phenology will be important in evaluating
the impact of climate change on maize yield. The response to
supra-optimal temperature differs substantially among the thermal
functions (Fig. 1).The process-based thermal functions TLU,
MAIZSIM, and EnzymResp are derived from data that includes
supra-optimal temperatures. These functions were derived from
controlled-environment studies during early phases of development
that showed that the rate of development declines rapidly when
temperatures are raised beyond the optimum temperature
(Lehenbauer, 1914; Tollenaar et al., 1979; Warrington and
Kanemasu, 1983a,b; Parent and Tardieu, 2012). In contrast, the
rate of development remains constant beyond 36C for empirical
linear functions and declines moderately beyond the optimum
temperature in nonlinear empirical functions that account for the
whole life cycle (Fig. 1). Nonlinear empirical functions were derived
based on data sets that contained either no or few supra-optimal
temperatures (i.e., Van Schaik and Probst, 1958; Brown, 1960;
Stewart et al., 1998), and the precision of these functions may be
low for supra-optimal conditions. Hence, while data are limited on
the supra-optimal range, we suggest that maize models that are used
to assess the impact of climate change on crop productivity should
utilize the developmental response to supra-optimal temperatures
of process-based functions for the pre-flowering period. Even less
is known about the developmental response to supra-optimal
temperatures during the grain-filling period. The GTI and CHU
functions performed best during the post-silking period in the
data sets evaluated here (Tables 3 and 4), with mean temperatures
<25C. To the best of our knowledge, no information is available
on the response of the grain-filling period to supra-optimal
temperatures. Owing to the paucity of supra-optimal temperature
data in the derivation or evaluation of empirical thermal functions,
process-based functions appear to better represent the temperature
response of maize development under supra-optimal temperatures.
Overestimation of the rate of development under supra-optimal
2096

temperatures by empirical models will lead to a disproportionate

reduction of the life cycle, and consequently, models that use
empirical thermal functions will underpredict yield in climatechange scenarios.
CONCLUSIONS
The results of this study show that large data sets can
improve the ability to differentiate among thermal functions
that quantify the effect of temperature on maize phenology.
The precision of nonlinear empirical functions, particularly
CHU, is superior to that of both linear empirical functions,
such as GDD10,30, and process-based functions, such as the
TLU function, in predicting maize phenology. The precision of
thermal functions was associated with the temperature response
across the 10 to 26C temperature range rather than with the
base and optimum cardinal temperatures or the supra-optimal
temperature range. The higher precision of the CHU function is
associated with its overall lower relative temperature sensitivity
in the 10 to 26C temperature range, which is probably due
to factors other than air temperature that influence rate of
development. The GTI function during the post-silking period
has a low temperature sensitivity for temperatures <23C (Fig.
1c), and this function was superior for the IndianaOhio data
set. Because the nonlinear empirical functions were developed
under conditions devoid of supra-optimal temperatures,
and while we await for research addressing the response to
temperature in that range, maize models that are used to assess
the impact of climate change on crop productivity should
utilize the developmental response of process-based functions
to quantify the response of phenology to supra-optimal
temperatures during the pre-flowering period.
REFERENCES
Abendroth, L.J., R.W. Elmore, M.J. Boyer, and S.K. Marlay. 2011. Corn growth
and development. PMR 1009. Iowa State Univ. Ext., Ames.
Asseng, S., F. Ewert, C. Rosenzweig, J.W. Jones, J. L. Hatfield, A.C. Ruane, et
al. 2013. Uncertainty in simulating wheat yields under climate change.
Nature Clim. Change 3:827832. doi:10.1038/nclimate1916
Birch, C.J., J. Vos, J. Kiniry, H.J. Bos, and A. Elings. 1998. Phyllochron responds
to acclimation to temperature and irradiance in maize. Field Crops Res.
59:187200. doi:10.1016/S0378-4290(98)00120-8
Blacklow, W.M. 1972. Influence of temperature on germination and elongation
of the radicle and shoot of corn (Zea mays L.). Crop Sci. 12:647650.
doi:10.2135/cropsci1972.0011183X001200050028x
Brown, D.M. 1960. Soybean ecology: I. Developmenttemperature relationships
from controlled-environment studies. Agron. J. 52:493496. doi:10.2134/
agronj1960.00021962005200090001x
Brown, D.M. 1969. Heat units for corn in southern Ontario. Ministry of Agric.
and Food, Guelph, ON, Canada.
Brown, D.M. 1977. Response of maize to environmental temperatures: A
review. In: W. Baier, editor, Proceedings of the WMO Symposium on
Agrometeorology of the Maize (Corn) Crop, Ames, IA. 59 July 1976.
World Meteorol. Organ., Geneva, Switzerland. p. 1526.
Brown, D.M., and A. Bootsma. 1993.Crop heat units for corn and other warm
season crops in Ontario. Ministry of Agriculture and Food, Guelph, ON,
Canada.
Cirilo, A.G., and F.H. Andrade. 1994a. Sowing date and maize productivity:
I. Crop growth and dry matter partitioning. Crop Sci. 34:10391043.
doi:10.2135/cropsci1994.0011183X003400040037x
Cirilo, A.G., and F.H. Andrade. 1994b. Sowing date and maize productivity: II.
Kernel number determination. Crop Sci. 34:10441046. doi:10.2135/cro
psci1994.0011183X003400040038x

Agronomy Journal Volume 106, Issue 6 2014

Edmeades, G.O., J. Bolaos, A. Elings, J.-M. Ribault, M. Bnziger, and M.E.

Westgate. 2000. The role and regulation of the anthesissilking interval
in maize. In: M. Westgate and K. Boote, editors, Physiology and modeling
kernel set in maize. CSSA Spec. Publ. 29. CSSA and ASA, Madison, WI.
p. 4373. doi:10.2135/cssaspecpub29.c4
Ellis, R.H., R.J. Summerfield, G.O. Edmeades, and E.H. Roberts. 1992.
Photoperiod, leaf number, and interval from tassel initiation to emergence
in diverse cultivars of maize. Crop Sci. 32:398403.
Gilmore, E.C., and J.S. Rogers. 1958. Heat units as a method of measuring
maturity in corn. Agron. J. 50:611615. doi:10.2134/agronj1958.000219
62005000100014x
Intergovernmental Panel on Climate Change. 2007. Climate change 2007:
Synthesis report. Contribution of Working Groups I, II, and III to the
Fourth Assessment Report of the Intergovernmental Panel on Climate
Change. IPCC, Geneva Switzerland. http://www.ipcc.ch/pdf/assessmentreport/ar4/syr/ar4_syr.pdf.
Johnson, F.H., H. Eyring, and R.W. Williams. 1942. The nature of enzyme
inhibitions in bacterial luminescence: Sulfanilamide, urethane,
temperature and pressure. J. Cell. Comp. Physiol. 20:247268.
doi:10.1002/jcp.1030200302
Jones, C.A., and J.R. Kiniry. 1986. CERES-Maize: A simulation model of maize
growth and development. Texas A&M Univ. Press, College Station, TX.
Kim, S.-H., Y. Yang, D.J. Timlin, D.H. Fleisher, A. Dathe, V.R. Reddy,
and K. Staver. 2012. Modeling temperature responses of leaf growth,
development, and biomass in maize with MAIZSIM. Agron. J. 104:1523
1537. doi:10.2134/agronj2011.0321
Kiniry, J.R., J.T. Ritchie, R.L. Musser, E.P. Flint, and W.C. Iwig. 1983.
The photoperiod sensitive interval in maize. Agron. J. 75:687690.
doi:10.2134/agronj1983.00021962007500040026x
Lauer, J. 1998. The Wisconsin comparative relative maturity (CRM) system
for corn. Field Crops 28.31-21. Agron. Advice (December). http://corn.
agronomy.wisc.edu/AA/A021.aspx.
Lehenbauer, P.A. 1914. Growth of maize seedlings in relation to temperature.
Physiol. Res. 1:247288.
Lizaso, J.I., K.J. Boote, J.W. Jones, C.H. Porter, L. Echarte, M.E. Westgate, and
G. Sonohat. 2011. CSM-IXIM: A new maize simulation model for DSSAT
Version 4.5. Agron. J. 103:766779. doi:10.2134/agronj2010.0423
McCullough, D.E., Ph. Girardin, M. Mihajlovic, A. Aguilera, and M. Tollenaar.
1994. Influence of N supply on development and dry matter accumulation
of an old and a new maize hybrid. Can. J. Plant Sci. 74:471477.
doi:10.4141/cjps94-087
Muldoon, J.F., T.B. Daynard, B. Van Duinen, and M. Tollenaar. 1984.
Comparisons among rates of appearance of leaf tips, collars, and leaf area
in maize (Zea mays L.). Maydica 29:109120.
Nielsen, R.L., P.R. Thomison, G.A. Brown, A.L. Halter, J. Wells, and
K.L. Wuethrich. 2002. Delayed planting effects on flowering and
grain maturation of dent corn. Agron. J. 94:549558. doi:10.2134/
agronj2002.5490
Parent, B., and F. Tardieu. 2012. Temperature responses of developmental
processes have not been affected by breeding in different
ecological areas for 17 crop species. New Phytol. 194:760774.
doi:10.1111/j.1469-8137.2012.04086.x

Parent, B., O. Turc, M. Gibon, M. Stitt, and F. Tardieu. 2010. Modelling

temperature-compensated physiological rates, based on the co-ordination
of responses to temperature of developmental processes. J. Exp. Bot.
61:20572069. doi:10.1093/jxb/erq003
R Development Core Team. 2013. R: A language and environment for statistical
computing. R Found. Stat. Comput., Vienna, Austria.
Sanchez, B.A., A. Rasmussen, and J.P. Porter. 2014. Temperatures and the
growth and development of maize and rice: A review. Global Change Biol.
20:408417. doi:10.1111/gcb.12389
Shaw, R.H., and H.C.S. Thom. 1951. On the phenology of field corn, silking to
maturity. Agron. J. 43:541546. doi:10.2134/agronj1951.000219620043
00110004x
Stewart, D.W., L.M. Dwyer, and L.L. Carrigan. 1998. Phenological temperature
response of maize. Agron. J. 90:7379. doi:10.2134/agronj1998.0002196
2009000010014x
Tollenaar, M. 1999. Duration of the grain-filling period in maize is not affected
by photoperiod and incident PPFD during the vegetative phase. Field
Crops Res. 62:1521. doi:10.1016/S0378-4290(98)00170-1
Tollenaar, M., T.B. Daynard, and R.B. Hunter. 1979. Effect of temperature on
rate of leaf appearance and flowering date in maize. Crop Sci. 19:363366.
doi:10.2135/cropsci1979.0011183X001900030022x
Tollenaar, M., and R.B. Hunter. 1983. A photoperiod and temperature sensitive
period for leaf number in maize. Crop Sci. 23:457460. doi:10.2135/crops
ci1983.0011183X002300030004x
Tollenaar, M., J.F. Muldoon, and T.B. Daynard. 1984. Differences in rate of leaf
appearance among maize hybrids and phases of development. Can. J. Plant
Sci. 64:759763. doi:10.4141/cjps84-104
Van Schaik, P.H., and A.H. Probst. 1958. Effects of some environmental factors
on flower production and reproductive efficiency in soybeans. Agron. J.
50:192197. doi:10.2134/agronj1958.00021962005000040007x
Vinocur, M.G., and J.T. Ritchie. 2001. Maize leaf development biases caused
by airapex temperature differences. Agron. J. 93:767772. doi:10.2134/
agronj2001.934767x
Warrington, I.J., and E.T. Kanemasu. 1983a. Corn growth response to
temperature and photoperiod: II. Leaf initiation and leaf appearance rates.
Agron. J. 75:755761. doi:10.2134/agronj1983.0002196200750005000
9x
Warrington, I.J., and E.T. Kanemasu. 1983b. Corn growth response to
temperature and photoperiod: III. Leaf number. Agron. J. 75:762766.
doi:10.2134/agronj1983.00021962007500050010x
Wilson, D.R., R.C. Muchow, and C.J. Murgatroyd. 1995. Model
analysis of temperature and solar radiation limitations to maize
potential productivity in a cool climate. Field Crops Res. 43:118.
doi:10.1016/0378-4290(95)00037-Q
Yan, W., and L.A. Hunt. 1999. An equation for modelling the temperature
response of plants using only the cardinal temperatures. Ann. Bot.
84:607614. doi:10.1006/anbo.1999.0955
Yang, H., A. Dobermann, D.T. Walters, T.J. Arkebauer, and K.G. Cassman.
2006. Hybrid-Maize: A simulation model for corn growth and yield. Dep.
of Agron. and Hortic., Univ. of Nebraska, Lincoln.
Yin, X., M.J. Kropff, G. MacLaren, and R.M. Visperas. 1995. A nonlinear model
for crop development as a function of temperature. Agric. For. Meteorol.
77:116. doi:10.1016/0168-1923(95)02236-Q

Agronomy Journal Volume 106, Issue 6 2014

2097